//! Terraform and OpenTofu state, kept by the hub per repository //! (oxygenhub#252): the HTTP backend Terraform and OpenTofu speak, with //! locking, at `/api/v1/repos/OWNER/NAME/terraform/state/STATE`, so a Flow //! job keeps its state without any cloud credential. A job authenticates //! with an identity token of this hub's for the audience `{base}/terraform` //! (its pipeline's `identity` block) or its runner's job token, either for //! its own repository only; people with a token of theirs and their //! repository role. Tokens go as the bearer, or as the password of Basic //! auth, which is what the backend sends. //! //! State is kept outside the repository's history (it often holds secrets), //! sealed at rest with the hub's secrets key, and its last versions are //! kept, so a bad apply can be rolled back: a kept version is restored as a //! new one (#255), so nothing is lost and going forward again is restoring //! a newer one. use crate::api::ApiError; use crate::auth::roles::{Access, Role}; use crate::{db, Shared}; use axum::body::Bytes; use axum::extract::{Path, Query, State}; use axum::http::{header, HeaderMap, Method, StatusCode}; use axum::response::{IntoResponse, Response}; /// Versions of a state kept besides the current one. pub const VERSIONS_KEPT: i64 = 20; /// The most a state may weigh. pub const MAX_STATE_BYTES: usize = 32 << 20; /// Who may use a repository's states. enum Who { /// A running job of this repository. Job { name: String }, /// A person, with their role on the repository. Person { account: Box, role: Role }, } impl Who { fn label(&self) -> String { match self { Who::Job { name } => name.clone(), Who::Person { account, .. } => account.username.clone(), } } fn may(&self, access: Access) -> bool { match self { // A job reads and writes its own repository's states. Who::Job { .. } => true, Who::Person { role, .. } => crate::auth::roles::allows(Some(*role), access), } } } fn err(status: StatusCode, why: impl Into) -> ApiError { ApiError(status, why.into()) } /// The token a request carries: the bearer, or Basic auth's password. fn token_of(headers: &HeaderMap) -> Option { if let Some(t) = oxy_server::bearer_token(headers) { return Some(t.to_string()); } let v = headers.get(header::AUTHORIZATION)?.to_str().ok()?; let b64 = v.strip_prefix("Basic ").or_else(|| v.strip_prefix("basic "))?; let raw = String::from_utf8(oxy_core::dsse::base64_decode(b64.trim())?).ok()?; let (_, password) = raw.split_once(':')?; (!password.is_empty()).then(|| password.to_string()) } /// The audience a job's identity token needs here. pub fn audience(base_url: &str) -> String { format!("{}/terraform", base_url.trim_end_matches('/')) } fn valid_name(name: &str) -> bool { !name.is_empty() && name.len() <= 100 && name.chars().all(|c| c.is_ascii_alphanumeric() || c == '-' || c == '_' || c == '.') && !name.starts_with('.') } async fn who(app: &Shared, headers: &HeaderMap, row: &db::repos::RepoRow) -> Result { let token = token_of(headers).ok_or_else(|| { err(StatusCode::UNAUTHORIZED, "a Terraform state needs a token: a job's (TF_HTTP_PASSWORD), or one of yours with repo:write") })?; let hash = crate::auth::token_hash(&token); if token.starts_with(crate::flow::JOB_TOKEN_PREFIX) { let job = db::flow::job_by_token(&app.db, &hash, crate::flow::JOB_TOKEN_MAX_AGE_MS) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "this job token is not valid (the job must be running)"))?; let run = db::flow::run_by_id(&app.db, job.run_id) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "this job token is not valid"))?; if run.repo_id != row.id { return Err(err(StatusCode::FORBIDDEN, "a job keeps the states of its own repository only")); } return Ok(Who::Job { name: format!("run #{} {}", run.number, job.name) }); } let personal = db::accounts::token_by_hash(&app.db, &hash).await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?; let Some(token) = personal.filter(|t| !t.expired(db::now_ms())) else { // A job's identity token for this hub's Terraform audience. let issuer = app.oidc().await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?; let aud = audience(app.config.base_url()); let Some(claims) = issuer.verify(&token, &aud, db::now_ms()) else { return Err(err(StatusCode::UNAUTHORIZED, format!("unknown or expired token (a job's identity token must be for {aud})"))); }; let sub = claims["sub"].as_str().unwrap_or(""); let repo = claims["repository"] .as_str() .map(String::from) .or_else(|| sub.strip_prefix("repo:").and_then(|r| r.split_once(":bookmark:")).map(|(r, _)| r.to_string())); if repo.as_deref() != Some(row.slug().as_str()) { return Err(err(StatusCode::FORBIDDEN, "a job keeps the states of its own repository only")); } return Ok(Who::Job { name: sub.to_string() }); }; if !(token.has_scope("repo:write") || token.has_scope("repo:admin")) { return Err(err( StatusCode::FORBIDDEN, "the token lacks repo:write: a Terraform state, read or written, needs it (states hold secrets)", )); } let slug = row.slug(); if !token.covers(&slug) { return Err(err(StatusCode::FORBIDDEN, format!("the token does not cover {slug}"))); } let account = db::accounts::account_by_id(&app.db, token.account_id) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .filter(|a| !a.banned && !a.disabled) .ok_or_else(|| err(StatusCode::UNAUTHORIZED, "unknown token"))?; let role = crate::auth::roles::role_for(&app.db, Some(&account), row) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?; match role { Some(r) if crate::auth::roles::allows(Some(r), Access::Read) => Ok(Who::Person { account: Box::new(account), role: r }), _ => Err(err(StatusCode::NOT_FOUND, "no such repository")), } } #[derive(serde::Deserialize, Default)] pub struct LockQuery { #[serde(rename = "ID", default)] id: Option, } fn json(status: StatusCode, body: String) -> Response { (status, [(header::CONTENT_TYPE, "application/json")], body).into_response() } /// The lock's ID in a lock-info body (`{"ID": …}`). fn lock_id(info: &[u8]) -> Option { serde_json::from_slice::(info).ok()?.get("ID")?.as_str().map(String::from) } /// Why a restore did not happen. pub enum RestoreError { /// No such version of that state. NotFound, /// Someone holds the state's lock: who. Locked(String), /// The kept body does not open or is not JSON. Broken, Db(sqlx::Error), } impl RestoreError { pub fn message(&self) -> String { match self { RestoreError::NotFound => "no such version of this state".into(), RestoreError::Locked(by) => format!("the state is locked by {by}: restore it once the lock is released"), RestoreError::Broken => "that version does not open: it cannot be restored".into(), RestoreError::Db(e) => e.to_string(), } } fn status(&self) -> StatusCode { match self { RestoreError::NotFound => StatusCode::NOT_FOUND, RestoreError::Locked(_) => StatusCode::LOCKED, RestoreError::Broken | RestoreError::Db(_) => StatusCode::INTERNAL_SERVER_ERROR, } } } /// Restore kept version `id` of a state (#255): written as a new version /// whose serial is above the current one (OpenTofu and Terraform take a /// state only forward), so the state it replaces is kept as a version too. /// Refused while the state is locked. The new serial. pub async fn restore(app: &Shared, repo_id: i64, state: &str, id: i64, by: &str) -> Result { let current = db::terraform::get(&app.db, repo_id, state).await.map_err(RestoreError::Db)?; if let Some(l) = current.as_ref().filter(|s| !s.lock_id.is_empty()) { return Err(RestoreError::Locked(l.locked_by.clone())); } let (body, old_serial) = db::terraform::version_body(&app.db, repo_id, state, id).await.map_err(RestoreError::Db)?.ok_or(RestoreError::NotFound)?; let plain = app.secret_key.open(&body).ok_or(RestoreError::Broken)?; let mut json: serde_json::Value = serde_json::from_str(&plain).map_err(|_| RestoreError::Broken)?; let serial = current.map_or(0, |c| c.serial).max(old_serial) + 1; json["serial"] = serial.into(); let text = serde_json::to_string_pretty(&json).map_err(|_| RestoreError::Broken)?; let label = format!("{by}, restoring serial {old_serial}"); db::terraform::put(&app.db, repo_id, state, &app.secret_key.seal(&text), serial, &label, VERSIONS_KEPT) .await .map_err(RestoreError::Db)?; Ok(serial) } /// `GET /api/v1/repos/{owner}/{name}/terraform/state/{state}/versions`: /// the state's kept versions, newest first (#255). pub async fn versions( State(app): State, headers: HeaderMap, Path((owner, name, state)): Path<(String, String, String)>, ) -> Result { let (row, _who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Write).await?; let list = db::terraform::version_list(&app.db, row.id, &state).await.map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))?; let out: Vec = list .iter() .map(|v| serde_json::json!({ "id": v.id, "serial": v.serial, "written_by": v.created_by, "written_ms": v.created_ms })) .collect(); Ok(json(StatusCode::OK, serde_json::Value::Array(out).to_string())) } /// `GET …/terraform/state/{state}/versions/{id}`: one kept version's state. pub async fn version( State(app): State, headers: HeaderMap, Path((owner, name, state, id)): Path<(String, String, String, i64)>, ) -> Result { let (row, _who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Write).await?; let (body, _) = db::terraform::version_body(&app.db, row.id, &state, id) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such version of this state"))?; let plain = app.secret_key.open(&body).ok_or_else(|| err(StatusCode::INTERNAL_SERVER_ERROR, "the version does not open"))?; Ok(json(StatusCode::OK, plain)) } /// `POST …/terraform/state/{state}/versions/{id}/restore`: restore a kept /// version, by a person with the maintain role (never a job), audited. pub async fn restore_version( State(app): State, headers: HeaderMap, Path((owner, name, state, id)): Path<(String, String, String, i64)>, ) -> Result { let (row, who) = versions_caller(&app, &headers, &owner, &name, &state, Access::Maintain).await?; let Who::Person { account, .. } = &who else { return Err(err(StatusCode::FORBIDDEN, "a job does not restore a state: a person with the maintain role does")); }; let serial = restore(&app, row.id, &state, id, &account.username).await.map_err(|e| err(e.status(), e.message()))?; db::accounts::audit( &app.db, Some(account), "terraform.restored", &format!("{}:{state}", row.slug()), &format!("version {id}, now serial {serial}"), ) .await .ok(); Ok(json(StatusCode::OK, serde_json::json!({ "serial": serial }).to_string())) } /// The repository and the caller of a versions endpoint, allowed `access`. async fn versions_caller( app: &Shared, headers: &HeaderMap, owner: &str, name: &str, state: &str, access: Access, ) -> Result<(db::repos::RepoRow, Who), ApiError> { if !valid_name(state) { return Err(err(StatusCode::BAD_REQUEST, "a state's name is letters, digits, '-', '_' and '.', up to 100")); } let row = db::repos::get(&app.db, owner, name) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such repository"))?; let who = who(app, headers, &row).await?; if !who.may(access) { return Err(err( StatusCode::FORBIDDEN, if access == Access::Maintain { "restoring a state needs the maintain role" } else { "a Terraform state needs the write role" }, )); } Ok((row, who)) } /// `GET|POST|DELETE|LOCK|UNLOCK /api/v1/repos/{owner}/{name}/terraform/state/{state}`. pub async fn state( State(app): State, method: Method, headers: HeaderMap, Path((owner, name, state)): Path<(String, String, String)>, Query(q): Query, body: Bytes, ) -> Result { if !valid_name(&state) { return Err(err(StatusCode::BAD_REQUEST, "a state's name is letters, digits, '-', '_' and '.', up to 100")); } let row = db::repos::get(&app.db, &owner, &name) .await .map_err(|e| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()))? .ok_or_else(|| err(StatusCode::NOT_FOUND, "no such repository"))?; let who = who(&app, &headers, &row).await?; // Reading a state needs the write role too: states hold secrets, and // every account reads a public or internal repository (#256). if !who.may(Access::Write) { return Err(err(StatusCode::FORBIDDEN, "a Terraform state needs the write role on its repository: states hold secrets")); } let db_err = |e: sqlx::Error| err(StatusCode::INTERNAL_SERVER_ERROR, e.to_string()); let current = db::terraform::get(&app.db, row.id, &state).await.map_err(db_err)?; let locked = current.as_ref().filter(|s| !s.lock_id.is_empty()); match method.as_str() { "GET" => match current.filter(|s| !s.body.is_empty()) { Some(s) => { let plain = app.secret_key.open(&s.body).ok_or_else(|| err(StatusCode::INTERNAL_SERVER_ERROR, "the state does not open"))?; Ok(json(StatusCode::OK, plain)) } // No state yet: Terraform starts afresh. None => Ok(StatusCode::NO_CONTENT.into_response()), }, "POST" => { if body.len() > MAX_STATE_BYTES { return Err(err(StatusCode::PAYLOAD_TOO_LARGE, format!("a state may be {} MiB at most", MAX_STATE_BYTES >> 20))); } let text = std::str::from_utf8(&body).map_err(|_| err(StatusCode::BAD_REQUEST, "a state is JSON"))?; let parsed: serde_json::Value = serde_json::from_str(text).map_err(|_| err(StatusCode::BAD_REQUEST, "a state is JSON"))?; if let Some(l) = locked { if q.id.as_deref() != Some(l.lock_id.as_str()) { return Ok(json(StatusCode::LOCKED, l.lock_info.clone())); } } let serial = parsed.get("serial").and_then(|v| v.as_i64()).unwrap_or(0); let sealed = app.secret_key.seal(text); db::terraform::put(&app.db, row.id, &state, &sealed, serial, &who.label(), VERSIONS_KEPT).await.map_err(db_err)?; Ok(StatusCode::OK.into_response()) } "DELETE" => { if let Some(l) = locked { if q.id.as_deref() != Some(l.lock_id.as_str()) { return Ok(json(StatusCode::LOCKED, l.lock_info.clone())); } } db::terraform::delete(&app.db, row.id, &state).await.map_err(db_err)?; Ok(StatusCode::OK.into_response()) } "LOCK" => { let info = std::str::from_utf8(&body).unwrap_or(""); let Some(id) = lock_id(body.as_ref()) else { return Err(err(StatusCode::BAD_REQUEST, "a lock request is the lock's JSON, with its ID")); }; if let Some(l) = locked { return Ok(json(if l.lock_id == id { StatusCode::OK } else { StatusCode::LOCKED }, l.lock_info.clone())); } if db::terraform::lock(&app.db, row.id, &state, &id, info, &who.label()).await.map_err(db_err)? { Ok(json(StatusCode::OK, info.to_string())) } else { // Taken meanwhile. let now = db::terraform::get(&app.db, row.id, &state).await.map_err(db_err)?; Ok(json(StatusCode::LOCKED, now.map(|s| s.lock_info).unwrap_or_default())) } } "UNLOCK" => { let Some(l) = locked else { return Ok(StatusCode::OK.into_response()) }; // `terraform force-unlock` sends the ID; a plain unlock its lock info. let id = lock_id(body.as_ref()).or(q.id.clone()); if id.as_deref() != Some(l.lock_id.as_str()) { return Ok(json(StatusCode::CONFLICT, l.lock_info.clone())); } db::terraform::unlock(&app.db, row.id, &state).await.map_err(db_err)?; Ok(StatusCode::OK.into_response()) } other => { Err(err(StatusCode::METHOD_NOT_ALLOWED, format!("{other} is not a Terraform backend method (GET, POST, DELETE, LOCK, UNLOCK)"))) } } }